Video Generation

Generate videos from text prompts or input images using Fluxpool's unified API. Multiple video models available — call GET /v1/models for the current catalog.

OpenAI-compatible. This endpoint follows the OpenAI images/generations pattern with video extensions. If you've used the OpenAI SDK, you're ready.

Overview#

Video generation is asynchronous. Video models typically take 30–120 seconds to produce output. The flow is:

Submit a generation request → receive an id and a presigned S3 URL
Poll GET /v1/generations/{id} (or the presigned URL) until the video is ready
Download the video from data[0].url before it expires (1 hour)

Note: Webhook delivery is on the roadmap. Until then, poll the generation-status endpoint every few seconds.

Endpoint#

POST https://api.fluxpool.ai/v1/videos/generations

Request Body#

Text-to-Video#

Generate a video from a text prompt.

JSON

{
  "model": "seedance-v2",
  "prompt": "A drone shot flying through cherry blossom trees, slow motion, cinematic lighting",
  "duration": 5,
  "resolution": "720p"
}

Image-to-Video#

Animate a static image into a video. Pass a reference image URL in the input_images array alongside (or instead of) prompt. The newer inputs field accepts the same URLs plus inline base64 data-URIs — see Image-to-Image for the shape.

JSON

{
  "model": "seedance-v2",
  "prompt": "Camera slowly zooms out, leaves rustling in wind",
  "input_images": ["https://your-bucket.s3.amazonaws.com/input-image.png"],
  "duration": 5,
  "resolution": "720p"
}

Not all models support image-to-video. Check the model's advertised capabilities via GET /v1/models.

Available Models#

Video models are discovered via GET /v1/models. One verified example to get you started:

Model ID	Notes
seedance-v2	Verified example. Confirmed working with resolution: "720p", duration: 5.

Other video model IDs are also live: seedance-v2.5 (Seedance 2.5), seedance-v2-fast (Seedance 2.0 Fast), seedance-v2-mini (Seedance 2.0 Mini), wan3.0-video-prime (Wan 3.0 Prime), wan3.0-video (Wan 3.0), wan2.7-t2v (Wan Video 2.7), wan2.7-r2v (Wan Video 2.7 · Reference), runway-gen4.5 (Runway Gen4.5). Full catalog: Browse all models →

Response#

A successful POST returns immediately with a generation id, a top-level status of "pending", and a data[] array holding a presigned S3 URL that becomes fetchable once the video is ready.

JSON — Initial response

{
  "created": 1785845810,
  "id": "6dcc89ce-964e-466e-8100-ba2bdfb6d07c",
  "status": "pending",
  "model": "seedance-v2",
  "type": "video",
  "data": [
    {
      "url": "https://fluxpool-output.s3.ap-southeast-1.amazonaws.com/...?X-Amz-...",
      "generation_id": "6dcc89ce-964e-466e-8100-ba2bdfb6d07c",
      "status": "pending",
      "model": "seedance-v2",
      "revised_prompt": null
    }
  ]
}

Possible status values:

Status	Description
pending	Generation is queued.
processing	Model is running.
completed	Video is ready — data[0].url is now fetchable.
failed	Generation failed. See error field.

The presigned URL is valid for 1 hour. Persist the video before it expires.

Polling for Results#

Poll the shared generation-status endpoint until status is "completed":

GET https://api.fluxpool.ai/v1/generations/{id}

Python

import time
import requests

API_KEY = "fp_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
headers = {"Authorization": f"Bearer {API_KEY}"}

# Submit
submit = requests.post(
    "https://api.fluxpool.ai/v1/videos/generations",
    headers=headers,
    json={"model": "seedance-v2", "prompt": "A koi fish transforming into a phoenix, slow motion", "resolution": "720p", "duration": 5}
).json()

gen_id = submit["id"]

# Poll
while True:
    status = requests.get(f"https://api.fluxpool.ai/v1/generations/{gen_id}", headers=headers).json()
    if status["status"] == "completed":
        print(status["data"][0]["url"])
        break
    time.sleep(3)

Polling interval: Wait at least 3 seconds between polls. Alternatively, poll the presigned S3 URL directly with GET and Range: bytes=0-0 (S3 rejects HEAD on presigned GETs).

Webhooks (coming soon)#

Webhook delivery is not shipped yet — polling is the supported pattern today. Track progress on the Async & polling page.

Code Examples#

# The OpenAI SDK doesn't have a videos helper — call the endpoint directly.
import requests

response = requests.post(
    "https://api.fluxpool.ai/v1/videos/generations",
    headers={"Authorization": "Bearer fp_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"},
    json={
        "model": "seedance-v2",
        "prompt": "A cyberpunk city street in heavy rain, neon reflections, cinematic",
        "duration": 5,
        "resolution": "720p"
    }
).json()

print(f"Generation ID: {response['id']}")
print(f"Poll URL: {response['data'][0]['url']}")

curl https://api.fluxpool.ai/v1/videos/generations \
  -H "Authorization: Bearer fp_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{"model": "seedance-v2", "prompt": "A cyberpunk city street in heavy rain, neon reflections, cinematic", "duration": 5, "resolution": "720p"}'

// The OpenAI SDK doesn't have a videos helper — use fetch.
const response = await fetch("https://api.fluxpool.ai/v1/videos/generations", {
  method: "POST",
  headers: {
    Authorization: "Bearer fp_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "seedance-v2",
    prompt: "A cyberpunk city street in heavy rain, neon reflections, cinematic",
    duration: 5,
    resolution: "720p",
  }),
});
const json = await response.json();
console.log(json.id, json.data[0].url);

Parameters Reference#

Parameter	Type	Required	Default	Description
model	string	Yes	—	Model ID. See Available Models.
prompt	string	Yes	—	Text description of the video to generate. Max 1000 characters.
inputs	object[]	No	null	Preferred shape. Each item is {type, source, value}. source: "url" for hosted images, source: "data" for base64 data-URIs (uploaded inline; ~4.5MB decoded cap per item).
input_images	string[]	No	null	Legacy. Prefer inputs. Array of https:// URLs for image-to-video workflows, kept for backwards compatibility.
duration	integer	No	5	Video duration in seconds. Range depends on model.
resolution	string	No	Model default	Output resolution. Verified values: "480p", "720p". Higher values (e.g. "1080p") are supported by some models.
parameters.seed	integer	No	Random	Seed for reproducibility when the underlying model supports seeding.
parameters.negative_prompt	string	No	null	What to avoid in the generation. Ignored by models that don't support it.

Error Handling#

If a generation fails, the response status will be "failed" with an error object:

JSON — Error Response

{
  "id": "gen_abc123xyz",
  "status": "failed",
  "error": { "code": "content_policy_violation", "message": "The prompt was flagged by the content safety filter." }
}

Code	Description
content_policy_violation	Prompt or output flagged by safety filters.
model_unavailable	The requested model is temporarily offline. Retry with an alternative model or contact support.
invalid_parameters	One or more parameters are invalid for the selected model.
insufficient_credits	Your account has insufficient credits. Top up →
CONCURRENCY_LIMIT_REACHED	HTTP 429. All of your plan's generations are running. Wait for one to finish. See Limits.
generation_timeout	The generation exceeded the maximum processing time. Retry.

Cost & Billing#

Each video generation deducts credits from your account balance. Cost varies by model, duration, and resolution.

Factor	Impact
Model	Higher-quality / premium provider models cost more per generation.
Duration	Longer videos cost more (roughly linear).
Resolution	Higher resolution costs more. 1080p > 720p.

Credits are deducted when generation begins. If a generation fails, credits are refunded automatically.

See full pricing details →

Limits#

Generations running at once. Each plan sets how many generations can be in progress at the same time for your account, whether they come from the app, the API or an MCP client. Without an active plan, your credits still work, at the lowest plan's limit.

Plan	Concurrent generations
1-month Trial	10
Base	10
Pro	15
Growth	20
Hero	25
Enterprise	Custom

Connected apps. An app or agent you connect through the MCP connector (OAuth) has its own daily credit limit, 500 by default, over a rolling 24 hours. Change it or pause the connection in the app under API / MCP Access, Connected apps. API keys have no daily limit.

Status	Code	When
429	CONCURRENCY_LIMIT_REACHED	Your plan's generations are all running. Wait for one to finish.
429	CONNECTION_LIMIT_REACHED	This connected app would go over its daily credit limit. The message says how much it has used and what the generation needs.
403	CONNECTION_PAUSED	This connected app is paused. Resume it to generate again.

Every error uses the same body. Branch on error.code; the message is for people and may change.

429 response body

{
  "error": {
    "message": "You have 10 generations running and your plan allows 10 at once. Wait for one to finish, or upgrade for more.",
    "type": "invalid_request_error",
    "param": null,
    "code": "CONCURRENCY_LIMIT_REACHED"
  }
}

Through the MCP server, the same object comes back as the text of the create_generation tool result. The API does not send rate-limit headers.

Need higher limits? Contact us →

📄

Image Generation

Generate images with 10+ models via the same API pattern.

🔔

Async & Polling

Polling patterns for long-running generations. Webhooks coming soon.

📋

Models Reference

Full parameter reference for all video models.

▶️

Try in Playground

Generate videos interactively — no code required.

Video Generation

Overview#

Endpoint#

Request Body#

Text-to-Video#

Image-to-Video#

Available Models#

Response#

Polling for Results#

Webhooks (coming soon)#

Code Examples#

Parameters Reference#

Error Handling#

Cost & Billing#

Limits#

Related#

Image Generation

Async & Polling

Models Reference

Try in Playground